Logan Kelly

Dr. Logan Kelly


Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
27 November 2021 in BMC Public Health
Verified
Journal Article
Open access
PDF file
AOAbdinardo M. B. de Oliveira
Abdinardo M. B. de Oliveira
JBJane Binner
Jane Binner
AMAnandadeep Mandal
Anandadeep Mandal
LKLogan Kelly
Logan Kelly
GPGabriel J. Power
Gabriel J. Power
Background: The COVID-19 pandemic has initiated several initiatives to better understand its behavior, and some projects are monitoring its evolution across countries, which naturally leads to comparisons made by those using the data. However, most “at a glance” comparisons may be misleading because the curve that should explain the evolution of COVID-19 is different across countries, as a result of the underlying geopolitical or socio-economic characteristics. Therefore, this paper contributes to the scientific endeavour by creating a new evaluation framework to help stakeholders adequately monitor and assess the evolution of COVID-19 in countries, considering the occurrence of spikes, "secondary waves" and structural breaks in the time series. Methods: Generalized Additive Models were used to model cumulative and daily curves for confirmed cases and deaths. The Root Relative Squared Error and the Percentage Deviance Explained measured how well the models fit the data. A local min-max function was used to identify all local maxima in the fitted values. The pure Markov-Switching and the family of Markov-Switching GARCH models were used to identify structural breaks in the COVID-19 time series. Finally, a quadrants system to identify countries that are more/less efficient in the short/long term in controlling the spread of the virus and the number of deaths was developed. Such methods were applied in the time series of 189 countries, collected from the Centre for Systems Science and Engineering at Johns Hopkins University. Results: Our methodology proves more effective in explaining the evolution of COVID-19 than growth functions worldwide, in addition to standardizing the entire estimation process in a single type of function. Besides, it highlights several inflection points and regime-switching moments, as a consequence of people’s diminished commitment to fighting the pandemic. Although Europe is the most developed continent in the world, it is home to most countries with an upward trend and considered inefficient, for confirmed cases and deaths. Conclusions: The new outcomes presented in this research will allow key stakeholders to check whether or not public policies and interventions in the fight against COVID-19 are having an effect, easily identifying examples of best practices and promote such policies more widely around the world.

Verified
Journal Article
Open access
PDF file
JBJane Binner
Jane Binner
LKLogan Kelly
Logan Kelly
Jonathan TepperJonathan Tepper
Jonathan Tepper
Accurate inflation forecasting is crucial for effective monetary policy, particularly during turning points that demand policy realignment. This study examines the efficacy of dedicating ensembles of shallow recurrent neural network models to different forecasting horizons for predicting U.S. inflation turning points more precisely than traditional methods, including the Survey of Professional Forecasters (SPF). We employ monthly data from January 1970 to May 2024, training these ensemble models on information through December 2022 and testing on out-of-sample observations from January 2023 to May 2024. The models generate forecasts at horizons of up to 16 months (one ensemble per horizon), accounting for both short- and medium-term dynamics. The results indicate that such ensembles of recurrent neural networks consistently outperform conventional approaches using key performance metrics, notably detecting inflation turning points earlier and projecting a return to target levels by May 2024—several months ahead of the Survey of Professional Forecasters’ average forecast. These findings underscore the value of such ensembles in capturing complex nonlinear relationships within macroeconomic data, offering a more robust alternative to standard econometric methods. By delivering timely and accurate forecasts, dedicated ensembles of shallow recurrent neural networks hold great promise for informing proactive policy measures and guiding decisions under uncertain economic conditions.

11 December 2024 in Preprints.org
Verified
Preprint
Open access
PDF file
JBJane Binner
Jane Binner
LKLogan Kelly
Logan Kelly
JTJonathan A. Tepper
Jonathan A. Tepper
Accurate inflation forecasting is critical for effective monetary policymaking, particularly during economic turning points that necessitate shifts in policy stance. This study evaluates the performance of multi-recurrent networks (MRNs), a class of recurrent neural networks with advanced memory mechanisms, in predicting inflation turning points more accurately than conventional methods such as the Survey of Professional Forecasters (SPF). Using U.S. macroeconomic data from January 1970 to May 2024, the MRNs were trained and validated on data through December 2022 and tested on out-of-sample data from January 2023 to May 2024. The models produced forecasts from t+1 to t+15 step-ahead, capturing both short- and medium-term inflation dynamics. Our findings show that MRNs, consistently outperform the SPF and other baseline models across key forecast evaluation metrics, including RMSE, MAE, and Theil’s U. Notably, the MRNs successfully identify inflation turning points earlier, predicting inflation to return to target levels by May 2024 - months ahead of SPF projections. This study underscores the utility of MRNs in capturing nonlinear and dynamic relationships within economic data, providing a robust alternative to traditional econometric approaches. These results highlight the potential of neural network-based methodologies to improve forecasting accuracy and inform more timely and effective policy decisions.

24 February 2025 in Preprints.org
Verified
Preprint
Open access
PDF file
JBJane Binner
Jane Binner
LKLogan Kelly
Logan Kelly
JTJonathan A. Tepper
Jonathan A. Tepper
Accurate inflation forecasting is crucial for effective monetary policy, particularly during turning points that demand policy realignment. This study examines the efficacy of Shallow Neural Network Ensemble Models in predicting U.S. inflation turning points more precisely than traditional methods, including the Survey of Professional Forecasters (SPF). We employ monthly data from January 1970 to May 2024, training these ensemble models on information through December 2022 and testing on out-of-sample observations from January 2023 to May 2024. The models generate forecasts at horizons of up to 16 months, accounting for both short- and medium-term dynamics. Results indicate that Shallow Neural Network Ensemble Models consistently outperform conventional approaches using key performance metrics, notably detecting inflation turning points earlier and projecting a return to target levels by May 2024 - several months ahead of the Survey of Professional Forecasters’ average forecast. These findings underscore the value of such ensembles in capturing complex nonlinear relationships within macroeconomic data, offering a more robust alternative to standard econometric methods. By delivering timely and accurate forecasts, Shallow Neural Network Ensemble Models hold great promise for informing proactive policy measures and guiding decisions under uncertain economic conditions.